VJ.PEAT: Automated measurement of prosodic features

Tillmann Pistor, Carsten Keil

This paper outlines the first steps of an innovative method of phonetic measurement of prosodic features, focusing on F0-slopes in local intonation patterns called PEAT. The technique presented is an algorithm aiming to measure phonetic differences in speech signals by applying machine-learning techniques. The process, operating on the basis of speech analysis and statistical computation programs such as Praat and R, successively uses robust acoustic variables processing, a smoothing process based on the physiology of natural articulation and extraction of compliant paths according to a generic cost function. This process allows a fully automated determination of the calibration parameters when conducting phonetic measurements with Praat, thus eliminating subjectivity and making the results and illustrations reliable and comparable. Furthermore, PEAT can be used to automatically detect, measure and classify prosodic units in unknown speech signals, thus bridging phonetics and machine-learning.

 DOI: 10.21437/SpeechProsody.2018-115

Cite as: Pistor, T., Keil, C. (2018) VJ.PEAT: Automated measurement of prosodic features. Proc. 9th International Conference on Speech Prosody 2018, 567-571, DOI: 10.21437/SpeechProsody.2018-115.

  author={Tillmann Pistor and Carsten Keil},
  title={VJ.PEAT: Automated measurement of prosodic features},
  booktitle={Proc. 9th International Conference on Speech Prosody 2018},